The abundance dynamics of short-lived marine species often exhibit nonlinear and noisy behavior, driven by unobserved species interactions. These complex dynamics pose challenges in forecasting and establishing robust reference points. Here, we introduce an empirical dynamic modeling (EDM) framework that addresses the challenges in constructing fishery benchmarks for short-lived species. We apply our framework to time series (>30 years) of federally managed brown shrimp stocks in the US Gulf of Mexico. Firstly, through simulations under various dynamics scenarios, we demonstrate the statistical robustness of EDM-based maximum sustainable yield (MSY). Secondly, we identify nonlinear signals and achieve high prediction accuracy in the empirical dynamics of brown shrimp, potentially contributing to the volatile fluctuations observed in annual abundance and fishery landings. Lastly, we obtain the EDM-based MSY of brown shrimp, providing fishery reference points urgently needed for timely and effective management. Our results highlight the utility of EDM in deriving reference points for short-lived species, particularly in situations where stock abundance is heavily influenced by unobserved specie interactions.